parity-suite

Run parity-suite with a feature slug and a current-side target after completing replace-strategy setup and the golden-dataset phase. It writes a Playwright suite with locator mapping, hand-written aria snapshots, and API characterization, then verifies strength via fault injection.

Write one Playwright suite against logical names resolved per side, and run it through projects with two baseURLs (current and new). The suite must be green on both sides; side-specific expectations are only allowed for declared intentional differences.

No. The design depends on Playwright-specific features: partial-match toMatchAriaSnapshot, role and accessible-name resolution via getByRole, and projects with two baseURLs. The Skill stops explicitly when Playwright is unavailable.

Mechanically diffing snapshots mixes genuine improvements (div becoming role=alert) with noise, producing no usable signal. Assertions must be hand-written tolerant snapshots covering only what the specification guarantees, with captured snapshots kept as reference material.

Fault injection is a negative control: known faults derived from regression categories are injected at observation boundaries, and the detection set (assertions plus differ) must turn red. A positive control run without injection must be green first, and passing faults reveal suite weaknesses.

It requires git, Node.js with Playwright, completed replace-strategy setup, and a finished golden-dataset phase A. The dataset's current target must match the selected target, and the suite runs one feature per execution.